- Dear all;
- As you know, Deep Learning (DL) techniques recently take a lot of attention in the different fields. For example, in the medical field the currently available annotated database online for the research purpose are not large enough to deal or handle via deep learning techniques. In contrast, deep learning techniques need a huge database to train very well and could apply further on medical database.
- Fortunately, recently many researchers present new idea to solve such this problem. They called that idea by "Transfer Learning". The main idea of transfer learning is train first deep learning model with a large annotated computer vision dataset such as ImageNet*, and then re-train or fine-tune the model utilizing the small annotated data that you have.
- The question is: What is the sense behind "Transfer Learning" that is recently utilized with Deep learning techniques when the database in the medical field is not large enough and there is no similarity between the nature of ImageNet database and the database from the medical field ?
------------------------------------------------------------------------------
*ImageNet is a huge set of natural images as in: http://www.image-net.org/
Or:
[1] O. Russakovsky et al., “Imagenet large scale visual recognition challenge,” Int. J. Comput. Vis., vol. 115, no. 3, pp. 211–252, 2015.
[2] Jia Deng, etal. Imagenet: alarge scale hierarchical image database,In:Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2009,pp.248–255.
- Badges
- Science topic
More Mugahed A. Al-antari's questions See All
Similar questions and discussions